Desktop research sources

Reddit, live (needs Clipper)

Fresh threads, subreddit listings and search, read by your own Chrome with your own login. Needs Clipper paired and the Reddit switch on. Make a free Reddit account and stay signed in to it in that Chrome: signed-in visits get far fewer human checks. Paced at about one page per second. Reddit refuses search requests made behind the scenes, so search needs Clipper 1.5.1 or later, which opens Reddit's normal search page in one background tab, scrolls it a few times slowly, reads the results and closes the tab. Until you have it, search with Reddit history.

Reddit history

Older posts and comments, searchable by keyword or subreddit and by date range. It comes from a public Reddit archive, so it needs no login and no Clipper. Use it to see how a complaint has moved over time.

Google Trends

Interest over time for up to 5 terms, and related queries for one term, by country and timeframe. Use it to check whether a problem is growing or fading. Google is touchy about speed, so Desktop waits about 4 to 5 seconds between requests and backs off if it is asked to slow down.

App Store and Google Play

App listings and reviews from both stores, by country. Search for apps in a niche, then pull the reviews of the ones that matter to see what users complain about. Apple allows about 20 searches a minute, so Desktop spaces them out.

Search suggestions

The autocomplete phrases people type, by country and language. Good for finding the real wording of a problem and long tail ideas.

YouTube

Search videos, read the comments under them, and read transcripts. For saved searches, outliers and comment themes, use the YouTube Lab. Comments under best-of and alternative videos are a good place to hear complaints. YouTube is not giving captions to apps right now, so a transcript is reported as unavailable. No account or key is needed.

Hacker News, Stack Exchange and GitHub

Hacker News search gives stories and comments with points and comment counts. Stack Exchange gives questions with views and answers, and a question with many views and no accepted answer is stuck demand. The free limit is 300 requests a day, and you can ask how many are left. GitHub search gives repositories and feature-request issues with reactions, at about 10 requests a minute, so it runs slowly.

Wayback pricing history

List the past captures of a page such as a competitor's pricing page, read a few of them, and see which prices and plan names changed. Pages that draw their prices with JavaScript show text changes only. Tracked pages, step-line timelines and a change log are in the Pricing History Lab.

App marketplaces

Shopify, WordPress.org and Atlassian. Fetch a category, rank its apps by users times how unhappy they are, then read the low-star reviews and see which apps get the most complaints about being too complicated. Pages are cached for a day. Watched categories, a landscape chart and a pricing ladder are in the App Marketplaces Lab.

Businesses for sale and Census counts

Listing cards from BizBuySell and BusinessesForSale. BusinessesForSale search pages work directly. BizBuySell and every detail page need Clipper with the Labs switch on. US Census county business counts need your own free Census key, which you paste in Settings. It is stored encrypted on your computer and never shown to the AI.

App Store keyword ranks

The App Store Lab checks keyword ranks in up to 60 countries. See the App Store Lab guide.

Review sites: G2, Capterra, TrustRadius and Trustpilot (needs Clipper)

Business buyers' reviews, read by your own Chrome through Clipper 1.5 or later, one page at a time, about every 9 seconds per site, with a daily limit. Each site is off until you switch it on in Settings, Connect Clipper, after a short notice. The Software Reviews Lab and Deep Research use them.

When a site's page does not match the app's reader, or a search page has no recognisable results, Desktop saves the page and your own AI reads it. A review is kept only if its text is on the page word for word, the rating is 1 to 5, nothing is invented and reviewer names are never stored. Desktop still chooses the pages: lowest rated first, star filters, newest. The Software Reviews Lab and the Deep Research plan show "N pages are waiting for your AI to read".

App Store and Google Play reviews, idea discovery, density and AI answers

The App Reviews Lab reads app reviews straight from both stores with no Clipper. Idea Discovery turns a category into big unhappy apps and idea cards. Local Density counts businesses by US state or county from the Census, with your own free key. AI Answers asks your own AI buyer questions and counts which names it recommends; its answers can differ for other people.

Labs

Several sources have a Lab: saved searches, charts, a sample that works offline, a weekly re-check while Desktop is open, and pins to the evidence board. Built Labs (all 15 are built): Trends, Reddit, Search Demand, Communities (Hacker News, Stack Exchange, GitHub), Businesses for Sale YouTube, App Marketplaces (Shopify, WordPress.org, Atlassian), Pricing History (Wayback Machine), Software Reviews (G2, Capterra, TrustRadius, Trustpilot), App Reviews (App Store, Google Play), Idea Discovery (category to big unhappy apps to idea cards), Local Density (US Census, needs your own free key), AI Answers (which names your own AI recommends) and the App Store Lab. To read one product across stores and forums in a single report, use Deep Research.

Any website, found by your AI (coming in a later Clipper update)

In Desktop, Settings, Connect Clipper, press Turn on Other sites and accept the short notice. From then on your own AI can search the web and read the pages it picks: Clipper runs the search on DuckDuckGo in a background tab of your own browser, your AI chooses the results worth reading, and Clipper opens each one, one page every 8 to 20 seconds. Desktop saves the readable text of each page as evidence in your project, with the site's name as the source and the address kept. Just ask in chat, for example: find and read 5 sites about invoice software for plumbers.

• No site list and no per-site prompt: Clipper 1.5 can read any public https page your AI asks for. Each site it reads shows in Settings as Found by your AI, with its own off switch and daily limit.
• You can still add specific addresses by hand in Add specific addresses (optional), one per line. To limit a site to part of it, write the site and a path with a star, such as example.com/blog/*.
• Deep Research uses it too: when Other sites is on, the Your sites lane asks your AI to find and read sites for the topic. You still approve the plan.
• Every page is one your AI or you asked for. Desktop never crawls or follows links by itself.
• If DuckDuckGo shows a check, the search stops with a clear message and is not retried.
• If a site shows a check or asks you to log in, Clipper stops and that site shows Needs you. Open the site, deal with it yourself, then press Resume in Clipper and in Desktop. Clipper never tries to get past a check.
• Addresses on your own computer or network are never opened. G2, Capterra, TrustRadius, Trustpilot and Reddit have their own routes and are not read here.
• A very long page is cut at 200,000 characters and marked as cut. Deep Research reads the first 30,000 characters of each page.
• Your own AI reads a saved page in chunks and saves facts with a quote. The app keeps a fact only when its quote is word for word in the saved page. The value is your AI's reading, and the app never makes up fields.

Some sites' terms limit automated access. Use it for your own research, keep the daily limits low, and switch a site or Other sites off at any time.

How pacing works

• Every request goes through a queue, one site at a time, with a set gap between calls.
• If a site says slow down, Desktop waits and retries instead of hammering it.
• Long jobs keep running in the background while you chat. Results are saved as evidence in the project.
• Reddit through Clipper is limited to one page a second so your account looks like normal browsing.